Random expert sampling for deep learning segmentation of acute ischemic stroke on non-contrast CT.

BACKGROUND: Outlining acutely infarcted tissue on non-contrast CT is a challenging task for which human inter-reader agreement is limited. We explored two different methods for training a supervised deep learning algorithm: one that used a segmentation defined by majority vote among experts and another that trained randomly on separate individual expert segmentations. METHODS: The data set consisted of 260 non-contrast CT studies in 233 patients with acute ischemic stroke recruited from the multicenter DEFUSE 3 (Endovascular Therapy Following Imaging Evaluation for Ischemic Stroke 3) trial. Additional external validation was performed using 33 patients with matched stroke onset times from the University Hospital Lausanne. A benchmark U-Net was trained on the reference annotations of three experienced neuroradiologists to segment ischemic brain tissue using majority vote and random expert sampling training schemes. The median of volume, overlap, and distance segmentation metrics were determined for agreement in lesion segmentations between (1) three experts, (2) the majority model and each expert, and (3) the random model and each expert. The two sided Wilcoxon signed rank test was used to compare performances (1) to 2) and (1) to (3). We further compared volumes with the 24 hour follow-up diffusion weighted imaging (DWI, final infarct core) and correlations with clinical outcome (modified Rankin Scale (mRS) at 90 days) with the Spearman method. RESULTS: The random model outperformed the inter-expert agreement ((1) to (2)) and the majority model ((1) to (3)) (dice 0.51±0.04 vs 0.36±0.05 (P<0.0001) vs 0.45±0.05 (P<0.0001)). The random model predicted volume correlated with clinical outcome (0.19, P<0.05), whereas the median expert volume and majority model volume did not. There was no significant difference when comparing the volume correlations between random model, median expert volume, and majority model to 24 hour follow-up DWI volume (P>0.05, n=51). CONCLUSION: The random model for ischemic injury delineation on non-contrast CT surpassed the inter-expert agreement ((1) to (2)) and the performance of the majority model ((1) to (3)). We showed that the random model volumetric measures of the model were consistent with 24 hour follow-up DWI.

Arterial-spin labeling MRI identifies residual cerebral arteriovenous malformation following stereotactic radiosurgery treatment.

BACKGROUND AND PURPOSE: Brain arteriovenous malformation (AVM) treatment by stereotactic radiosurgery (SRS) is effective, but AVM obliteration following SRS may take two years or longer. MRI with arterial-spin labeling (ASL) may detect brain AVMs with high sensitivity. We determined whether brain MRI with ASL may accurately detect residual AVM following SRS treatment. MATERIALS AND METHODS: We performed a retrospective cohort study of patients who underwent brain AVM evaluation by DSA between June 2010 and June 2015. Inclusion criteria were: (1) AVM treatment by SRS, (2) follow-up MRI with ASL at least 30 months after SRS, (3) DSA within 3 months of the follow-up MRI with ASL, and (4) no intervening AVM treatment between the MRI and DSA. Four neuroradiologists blindly and independently reviewed follow-up MRIs. Primary outcome measure was residual AVM indicated by abnormal venous ASL signal. RESULTS: 15 patients (12 females, mean age 29 years) met inclusion criteria. There were three posterior fossa AVMs and 12 supratentorial AVMs. Spetzler-Martin (SM) Grades were: SM1 (8%), SM2 (33%), SM3 (17%), SM4 (25%), and SM5 (17%). DSA demonstrated residual AVM in 10 patients. The pooled sensitivity, specificity, positive predictive value, and negative predictive value of venous ASL signal for predicting residual AVM were 100% (95% CI: 0.9-1.0), 95% (95% CI: 0.7-1.0), 98% (95% CI: 0.9-1.0), and 100% (95% CI: 0.8-1.0), respectively. High inter-reader agreement as found by Fleiss' Kappa analysis (k = 0.92; 95% CI: 0.8-1.0; P < 0.0001). CONCLUSIONS: ASL is highly sensitive and specific in the detection of residual cerebral AVM following SRS treatment.

Time From Imaging to Endovascular Reperfusion Predicts Outcome in Acute Stroke.

BACKGROUND AND PURPOSE: This study aims to describe the relationship between computed tomographic (CT) perfusion (CTP)-to-reperfusion time and clinical and radiological outcomes, in a cohort of patients who achieve successful reperfusion for acute ischemic stroke. METHODS: We included data from the CRISP (Computed Tomographic Perfusion to Predict Response in Ischemic Stroke Project) in which all patients underwent a baseline CTP scan before endovascular therapy. Patients were included if they had a mismatch on their baseline CTP scan and achieved successful endovascular reperfusion. Patients with mismatch were categorized into target mismatch and malignant mismatch profiles, according to the volume of their Tmax >10s lesion volume (target mismatch, <100 mL; malignant mismatch, >100 mL). We investigated the impact of CTP-to-reperfusion times on probability of achieving functional independence (modified Rankin Scale, 0-2) at day 90 and radiographic outcomes at day 5. RESULTS: Of 156 included patients, 108 (59%) had the target mismatch profile, and 48 (26%) had the malignant mismatch profile. In patients with the target mismatch profile, CTP-to-reperfusion time showed no association with functional independence (P=0.84), whereas in patients with malignant mismatch profile, CTP-to-reperfusion time was strongly associated with lower probability of functional independence (odds ratio, 0.08; P=0.003). Compared with patients with target mismatch, those with the malignant mismatch profile had significantly more infarct growth (90 [49-166] versus 43 [18-81] mL; P=0.006) and larger final infarct volumes (110 [61-155] versus 48 [21-99] mL; P=0.001). CONCLUSIONS: Compared with target mismatch patients, those with the malignant profile experience faster infarct growth and a steeper decline in the odds of functional independence, with longer delays between baseline imaging and reperfusion. However, this does not exclude the possibility of treatment benefit in patients with a malignant profile.

Can diffusion- and perfusion-weighted imaging alone accurately triage anterior circulation acute ischemic stroke patients to endovascular therapy?

BACKGROUND AND PURPOSE: Acute ischemic stroke (AIS) patients who benefit from endovascular treatment have a large vessel occlusion (LVO), small core infarction, and salvageable brain. We determined if diffusion-weighted imaging (DWI) and perfusion-weighted imaging (PWI) alone can correctly identify and localize anterior circulation LVO and accurately triage patients to endovascular thrombectomy (ET). MATERIALS AND METHODS: This retrospective cohort study included patients undergoing MRI for the evaluation of AIS symptoms. DWI and PWI images alone were anonymized and scored for cerebral infarction, LVO presence and LVO location, DWI-PWI mismatch, and ET candidacy. Readers were blinded to clinical data. The primary outcome measure was accurate ET triage. Secondary outcomes were detection of LVO and LVO location. RESULTS: Two hundred and nineteen patients were included. Seventy-three patients (33%) underwent endovascular AIS treatment. Readers correctly and concordantly triaged 70 of 73 patients (96%) to ET (κ=0.938; P=0.855) and correctly excluded 143 of 146 patients (98%; P=0.942). DWI and PWI alone had a 95.9% sensitivity and a 98.4% specificity for accurate endovascular triage. LVO were accurately localized to the ICA/M1 segment in 65 of 68 patients (96%; κ=0.922; P=0.817) and the M2 segment in 18 of 20 patients (90%; κ=0.830; P=0.529). CONCLUSION: AIS patients with anterior circulation LVO are accurately identified using DWI and PWI alone, and LVO location may be correctly inferred from PWI. MRA omission may be considered to expedite AIS triage in hyperacute scenarios or may confidently supplant non-diagnostic or artifact-limited MRA.